BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20121114T214500Z
DTEND:20121114T220000Z
LOCATION:155-F
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: While GPUs are computational powerhouses, GPU clusters are largely inefficient due to multiple data transfer costs across the PCIe bus. I have developed MPI-ACC, a high performance communication library for end-to-end data movement in CPU-GPU systems, where MPI-ACC is an extension to the popular MPI parallel programming paradigm. I provide a wide range of optimizations for point-to-point communication within MPI-ACC, which can be seamlessly leveraged by the application developers. I also show how MPI-ACC can further enable new application-specific optimizations, like efficient CPU-GPU co-scheduling and simultaneous CPU-GPU computation and network-GPU communication for improved system efficiency. I have also developed performance models to predict realistic performance bounds for GPU kernels, and this knowledge is used for optimal task distribution between the CPUs and GPUs for better efficiency. =0ALastly, I define a general efficiency metric for heterogeneous computing systems and show how MPI-ACC improves the overall efficiency of CPU-GPU based heterogeneous systems.
SUMMARY:Programming High Performance Heterogeneous Computing Systems: Paradigms, Models and Metrics
PRIORITY:3
END:VEVENT
END:VCALENDAR